Summary

Local designs the guide; only remote publishes it.

Both paths build and export the guide; only the hosted one deploys it and hands you a link a participant can open. Since the whole value of a survey is that someone answers it, the remote server is effectively the product and the local install is a design tool. The AI probing is the reason to prefer this over a form builder: a follow-up question written in response to what someone actually said is where the useful answers come from.

What it is

An MCP server for Feedbk, a survey product whose interviews are moderated by an AI that asks its own follow-up questions. You describe the research you want in plain language and the server turns it into a configured interview guide, detecting the right question type for each question rather than making you pick one.

What you get

Interview design from a description — name, purpose and questions — with the question type detected and configured automatically
AI probing built into the interview itself: the interviewer asks intelligent follow-ups based on what a participant answers, with nothing to configure
The question types you would expect, ready to use: single-choice, multiple-choice, open-ended text, and a rating scale from very dissatisfied to very satisfied
Guide export, available in both the local and remote setups
**Two capabilities the remote server has and the local install does not**: automatic deployment to feedbk.ai, and shareable interview links you can send to participants. A local install can create and export guides but cannot publish them

Requirements

Nothing to install for the remote path — the server is at `https://mcp.feedbk.ai/mcp` over HTTP, added with `claude mcp add --transport http feedbk https://mcp.feedbk.ai/mcp`, `gemini mcp add feedbk --url ...`, `codex mcp add feedbk --url ...`, as a VS Code HTTP server entry, or as a Claude Desktop custom connector. ChatGPT support needs a Pro, Team, Enterprise or Edu subscription and Developer Mode enabled under advanced settings. A local install is Python 3.10 or newer with the `feedbk-skill-mcp` package, and gives up deployment and shareable links. MIT licensed.

Setup effort

One command — claude mcp add --transport http feedbk https://mcp.feedbk.ai/mcp